Abstract:In sub‐Saharan Africa, acute‐onset severe malaria anaemia (SMA) is a life‐threatening condition, particularly affecting children under five, requiring immediate blood transfusion followed by the administration of antimalarial treatment. The WHO defines SMA as a haemoglobin (Hb) concentration lower than 0.5 g/dL (or a haematocrit <16.0%) in the presence of any parasite density. The acute drop in haematocrit in SMA is thought to be driven by an increased phagocytotic pathological process in the spleen, leading to the presence of distinct red blood cells (RBCs) with altered morphological characteristics. While point‐of‐care diagnosis is based on clinical presentation at admission, assessment of haematocrit and the analysis of a peripheral thick blood film to detect the presence of malaria parasites, SMA is challenging to diagnose in patients presenting with very low parasitaemia. In this study, we develop and validate a novel deep learning tool (Multiple Instance Learning framework, to Identify SMA [MILISMA]) designed to complement existing SMA diagnostic frameworks by identifying subtle, morphological changes in RBCs that potentially signal the onset of SMA prior to the manifestation of clinical symptoms as defined by current WHO criteria. Our model can systematically and at scale discern these early RBC morphological signals, thereby offering a valuable resource for closely monitoring patients at risk of SMA. The presence of such abnormal RBCs in high numbers could be a clinical indicator that a patient is developing SMA, and it is especially useful in very low parasitaemia cases. Such predictive capability is particularly crucial in facilitating timely interventions, potentially preventing the progression to severe anaemia and thus improving patient outcomes. Summary In sub‐Saharan Africa, acute‐onset severe malaria anaemia (SMA) is a critical challenge, particularly affecting children under five. The acute drop in haematocrit in SMA is thought to be driven by an increased phagocytotic pathological process in the spleen, leading to the presence of distinct red blood cells (RBCs) with altered morphological characteristics. We hypothesized that these RBCs could be detected systematically and at scale in peripheral blood films (PBFs) by harnessing the capabilities of deep learning models. Assessment of PBFs by a microscopist does not scale for this task and is subject to variability. Here we introduce a deep learning model, leveraging a weakly supervised Multiple Instance Learning framework, to Identify SMA (MILISMA) through the presence of morphologically changed RBCs. MILISMA achieved a classification accuracy of 83% (receiver operating characteristic area under the curve [AUC] of 87%; precision‐recall AUC of 76%). More importantly, MILISMA's capabilities extend to identifying statistically significant morphological distinctions (p

Leveraging deep learning for detecting red blood cell morphological changes in blood films from children with severe malaria anaemia

Advancing Malaria Identification From Microscopic Blood Smears Using Hybrid Deep Learning Frameworks

Localization of Malaria Parasites and White Blood Cells in Thick Blood Smears

Deep Learning-Based Cell Detection and Extraction in Thin Blood Smears for Malaria Diagnosis.

Efficient deep learning-based approach for malaria detection using red blood cell smears

Classification of Red Blood Cell Abnormality in Thin Blood Smear Images using Convolutional Neural Networks

Biophysical Profiling of Red Blood Cells from Thin-film Blood Smears using Deep Learning

An optimised YOLOv4 deep learning model for efficient malarial cell detection in thin blood smear images

Automated Detection of Acute Promyelocytic Leukemia in Blood Films and Bone Marrow Aspirates with Annotation-free Deep Learning

Clustering-Based Dual Deep Learning Architecture for Detecting Red Blood Cells in Malaria Diagnostic Smears.

An Efficient Deep Learning Approach for Malaria Parasite Detection in Microscopic Images

Fully-automated patient-level malaria assessment on field-prepared thin blood film microscopy images, including Supplementary Information

Pre-trained convolutional neural networks as feature extractors toward improved malaria parasite detection in thin blood smear images

Simulating Malaria Detection in Laboratories using Deep Learning

Detection of Peripheral Malarial Parasites in Blood Smears Using Deep Learning Models

Malaria Parasite Detection on Microscopic Blood Smear Images with Integrated Deep Learning Algorithms

Improving Malaria Parasite Detection from Red Blood Cell using Deep Convolutional Neural Networks

Deep Learning Enhanced Extended Depth-of-Field for Thick Blood-Film Malaria High-Throughput Microscopy

Ensembling Object Detection Models for Robust and Reliable Malaria Parasite Detection in Thin Blood Smear Microscopic Images

Malaria cell identification using improved machine learning and modified deep learning architecture

Residual Attention Learning Network and SVM for Malaria Parasite Detection